diff --git a/CMakeLists.txt b/CMakeLists.txt
index 06c0cff61..585314c27 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -28,7 +28,7 @@ set(PROJECT_VERSION_BUILD dev)
 
 # Append local cmake module path
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_LIST_DIR}/cpp/cmake/modules")
-project(libcucim VERSION ${VERSION} DESCRIPTION "libcucim" LANGUAGES CXX CUDA)
+project(libcucim VERSION ${VERSION} DESCRIPTION "libcucim" LANGUAGES CXX)
 
 ################################################################################
 # Include utilities
@@ -62,19 +62,7 @@ endif ()
 
 find_package(CUDAToolkit REQUIRED)
 set(CMAKE_CXX_STANDARD 17)
-set(CMAKE_CUDA_STANDARD 17) # Clion issue: https://youtrack.jetbrains.com/issue/CPP-19165 (fixed)
-set(CMAKE_CUDA_STANDARD_REQUIRED YES)
 set(CMAKE_CXX_STANDARD_REQUIRED YES)
-cucim_define_cuda_architectures(60;70;75;80;86)
-# https://github.com/Kitware/CMake/blob/master/Modules/Compiler/NVIDIA-CUDA.cmake#L11
-# https://gitlab.kitware.com/cmake/cmake/-/issues/19017
-# For CUDA >= 10.2, we cannot use --compiler-options as '-forward-unknown-to-host-compiler' wouldbe added by default to nvcc options.
-# For the reason, we add "${CMAKE_CXX_FLAGS}" instead of "--compiler-options ${CMAKE_CXX_FLAGS}" here.
-# ==> We changed to use "${CMAKE_CUDA_FLAGS}" instead of "${CMAKE_CXX_FLAGS}" ${CMAKE_CXX_FLAGS} can have wrong options such as '-march=nocona'
-set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -use_fast_math -Xptxas=-v")
-set(CMAKE_CUDA_FLAGS_DEBUG "${CMAKE_CUDA_FLAGS_DEBUG} -G")
-set(CMAKE_CUDA_FLAGS_RELEASE "${CMAKE_CUDA_FLAGS_RELEASE} -lineinfo")
-set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "${CMAKE_CUDA_FLAGS_RELWITHDEBINFO} -lineinfo")
 
 # Include CUDA headers explicitly for VSCode intelli-sense
 include_directories(AFTER SYSTEM ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
@@ -113,14 +101,23 @@ option(CUCIM_SUPPORT_NVTX "Support NVTX" ON)
 #   whether added before or after this command is invoked, and for the ones in sub-directories added after.
 add_definitions(-D_GLIBCXX_USE_CXX11_ABI=0) # TODO: create two library, one with CXX11 ABI and one without it.
 
+################################################################################
+# Define basic dependencies
+################################################################################
+# Include pthread
+# (https://cmake.org/cmake/help/v3.18/module/FindThreads.html)
+set(CMAKE_THREAD_PREFER_PTHREAD TRUE)
+set(THREADS_PREFER_PTHREAD_FLAG TRUE)
+find_package(Threads REQUIRED)
+
 ################################################################################
 # Define dependencies
 ################################################################################
 superbuild_depend(fmt)
 #superbuild_depend(boost)
 superbuild_depend(abseil)
-superbuild_depend(rmm) # this imports googletest internally
-#superbuild_depend(googletest)
+# superbuild_depend(rmm) # this imports googletest internally
+superbuild_depend(googletest)
 superbuild_depend(googlebenchmark)
 superbuild_depend(openslide)
 superbuild_depend(catch2)
@@ -156,13 +153,8 @@ configure_file(${CMAKE_CURRENT_LIST_DIR}/examples/cpp/CMakeLists.txt.examples.re
 ################################################################################
 set(INSTALL_TARGETS
         ${CUCIM_PACKAGE_NAME}
-#        ${CUCIM_PACKAGE_NAME}-header-only
-#        rmm
         fmt-header-only
-#        spdlog_header_only # required by rmm
         cucim_benchmarks
-#        cufile_stub
-#        cucim_tests
         )
 
 install(TARGETS ${INSTALL_TARGETS}
diff --git a/benchmarks/CMakeLists.txt b/benchmarks/CMakeLists.txt
index d7f9edcf1..0b6fa8d42 100644
--- a/benchmarks/CMakeLists.txt
+++ b/benchmarks/CMakeLists.txt
@@ -18,18 +18,12 @@
 ################################################################################
 
 add_executable(cucim_benchmarks main.cpp config.h)
-#set_source_files_properties(main.cpp PROPERTIES LANGUAGE CUDA) # failed with CLI11 library
 
 set_target_properties(cucim_benchmarks
     PROPERTIES
         CXX_STANDARD 17
         CXX_STANDARD_REQUIRED YES
         CXX_EXTENSIONS NO
-        CUDA_STANDARD 17
-        CUDA_STANDARD_REQUIRED YES
-        CUDA_EXTENSIONS NO
-        CUDA_SEPARABLE_COMPILATION ON
-        CUDA_RUNTIME_LIBRARY Shared
 )
 target_compile_features(cucim_benchmarks PRIVATE ${CUCIM_REQUIRED_FEATURES})
 # Use generator expression to avoid `nvcc fatal   : Value '-std=c++17' is not defined for option 'Werror'`
@@ -63,11 +57,6 @@ set_target_properties(cucim_primitives_benchmarks
         CXX_STANDARD 17
         CXX_STANDARD_REQUIRED YES
         CXX_EXTENSIONS NO
-        CUDA_STANDARD 17
-        CUDA_STANDARD_REQUIRED YES
-        CUDA_EXTENSIONS NO
-        CUDA_SEPARABLE_COMPILATION ON
-        CUDA_RUNTIME_LIBRARY Shared
 )
 target_compile_features(cucim_primitives_benchmarks PRIVATE ${CUCIM_REQUIRED_FEATURES})
 # Use generator expression to avoid `nvcc fatal   : Value '-std=c++17' is not defined for option 'Werror'`
diff --git a/cpp/CMakeLists.txt b/cpp/CMakeLists.txt
index fc139eb23..f6670c982 100644
--- a/cpp/CMakeLists.txt
+++ b/cpp/CMakeLists.txt
@@ -100,7 +100,7 @@ add_library(${CUCIM_PACKAGE_NAME}
         src/io/format/image_format.cpp
         src/logger/logger.cpp
         src/logger/timer.cpp
-        src/memory/memory_manager.cu
+        src/memory/memory_manager.cpp
         src/plugin/image_format.cpp
         src/plugin/plugin_config.cpp
         src/profiler/profiler.cpp
@@ -114,17 +114,12 @@ set_target_properties(${CUCIM_PACKAGE_NAME}
         CXX_STANDARD 17
         CXX_STANDARD_REQUIRED YES
         CXX_EXTENSIONS NO
-        CUDA_STANDARD 17
-        CUDA_STANDARD_REQUIRED YES
-        CUDA_EXTENSIONS NO
-        CUDA_SEPARABLE_COMPILATION ON
-        CUDA_RUNTIME_LIBRARY Shared
         SOVERSION ${PROJECT_VERSION_MAJOR}
         VERSION ${PROJECT_VERSION}
 )
 
 # Note: Looks like the following line causes error on CMake 3.18.4 (it is working on 3.18.2). Keeping it for now.
-set(CUCIM_REQUIRED_FEATURES cxx_std_17 cuda_std_17)
+set(CUCIM_REQUIRED_FEATURES cxx_std_17)
 target_compile_features(${CUCIM_PACKAGE_NAME} PRIVATE ${CUCIM_REQUIRED_FEATURES})
 # Use generator expression to avoid `nvcc fatal   : Value '-std=c++17' is not defined for option 'Werror'`
 target_compile_options(${CUCIM_PACKAGE_NAME}
@@ -150,13 +145,7 @@ target_link_libraries(${CUCIM_PACKAGE_NAME}
         PUBLIC
             ${CMAKE_DL_LIBS}
             $<BUILD_INTERFACE:deps::fmt>
-#            $<BUILD_INTERFACE:deps::boost>
-#            $<BUILD_INTERFACE:deps::rmm>
-#            $<BUILD_INTERFACE:deps::gds>
             $<INSTALL_INTERFACE:cucim::fmt-header-only>
-##            $<INSTALL_INTERFACE:cucim::boost>
-            # $<INSTALL_INTERFACE:cucim::rmm>
-##            $<INSTALL_INTERFACE:cucim::gds>
         PRIVATE
             CUDA::cudart
             deps::abseil
@@ -179,8 +168,6 @@ target_include_directories(${CUCIM_PACKAGE_NAME}
         PUBLIC
             $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>
             $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include/${CUCIM_PACKAGE_NAME}/3rdparty>
-#            $<BUILD_INTERFACE:${CMAKE_SOURCE_DIR}/gds/lib> # for GDS header
-#            $<BUILD_INTERFACE:${THRUST_INCLUDE_DIR}>
             $<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}>
             $<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/${CUCIM_PACKAGE_NAME}/3rdparty> # for 3rdparty libraries such as dlpack, nvtx3, and fmt
         PRIVATE
diff --git a/cpp/cmake/deps/nvtx3.cmake b/cpp/cmake/deps/nvtx3.cmake
index bc56c6ca8..f079181ab 100644
--- a/cpp/cmake/deps/nvtx3.cmake
+++ b/cpp/cmake/deps/nvtx3.cmake
@@ -30,6 +30,9 @@ if (NOT TARGET deps::nvtx3)
     # Create shared library
     cucim_set_build_shared_libs(ON) # since nvtx3 is header-only library, this may not needed.
 
+    set(BUILD_TESTS OFF)
+    set(BUILD_BENCHMARKS OFF)
+
     add_subdirectory(${deps-nvtx3_SOURCE_DIR}/cpp ${deps-nvtx3_BINARY_DIR} EXCLUDE_FROM_ALL)
     cucim_restore_build_shared_libs()
 
diff --git a/cpp/include/cucim/io/device.h b/cpp/include/cucim/io/device.h
index c1bd03784..7e1e5c7cc 100644
--- a/cpp/include/cucim/io/device.h
+++ b/cpp/include/cucim/io/device.h
@@ -38,6 +38,7 @@ class EXPORT_VISIBLE Device
 public:
     explicit Device();
     Device(const Device& device);
+    Device& operator=(const Device& device) = default;
     explicit Device(const std::string& device_name);
     Device(const char* device_name);
     Device(DeviceType type, DeviceIndex index);
diff --git a/cpp/include/cucim/memory/memory_manager.h b/cpp/include/cucim/memory/memory_manager.h
index d2246e9e1..47a567207 100644
--- a/cpp/include/cucim/memory/memory_manager.h
+++ b/cpp/include/cucim/memory/memory_manager.h
@@ -77,7 +77,7 @@ CUCIM_API void get_pointer_attributes(PointerAttributes& attr, const void* ptr);
  * @param dst_device Destination device of the memory.
  * @return `true` if succeed.
  */
-bool move_raster_from_host(void** target, size_t size, const cucim::io::Device& dst_device);
+CUCIM_API bool move_raster_from_host(void** target, size_t size, const cucim::io::Device& dst_device);
 
 /**
  * @brief Move device memory of `size` bytes to a new memory in `out_device`.
@@ -90,7 +90,7 @@ bool move_raster_from_host(void** target, size_t size, const cucim::io::Device&
  * @param dst_device Destination device of the memory.
  * @return `true` if succeed.
  */
-bool move_raster_from_device(void** target, size_t size, const cucim::io::Device& dst_device);
+CUCIM_API bool move_raster_from_device(void** target, size_t size, const cucim::io::Device& dst_device);
 
 } // namespace cucim::memory
 #endif // CUCIM_MEMORY_MANAGER_H
diff --git a/cpp/plugins/cucim.kit.cumed/CMakeLists.txt b/cpp/plugins/cucim.kit.cumed/CMakeLists.txt
index 1415ade8d..3917bb571 100644
--- a/cpp/plugins/cucim.kit.cumed/CMakeLists.txt
+++ b/cpp/plugins/cucim.kit.cumed/CMakeLists.txt
@@ -28,7 +28,7 @@ file(STRINGS ${CMAKE_CURRENT_LIST_DIR}/VERSION VERSION)
 # Append local cmake module path
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_LIST_DIR}/cmake/modules")
 
-project(cumed VERSION ${VERSION} DESCRIPTION "cumed" LANGUAGES CXX CUDA)
+project(cumed VERSION ${VERSION} DESCRIPTION "cumed" LANGUAGES CXX)
 set(CUCIM_PLUGIN_NAME "cucim.kit.cumed")
 
 ################################################################################
@@ -72,19 +72,7 @@ endif()
 find_package(CUDAToolkit REQUIRED)
 
 set(CMAKE_CXX_STANDARD 17)
-set(CMAKE_CUDA_STANDARD 17) # Clion issue: https://youtrack.jetbrains.com/issue/CPP-19165 (fixed)
-set(CMAKE_CUDA_STANDARD_REQUIRED YES)
 set(CMAKE_CXX_STANDARD_REQUIRED YES)
-cucim_define_cuda_architectures(60;70;75;80;86)
-# https://github.com/Kitware/CMake/blob/master/Modules/Compiler/NVIDIA-CUDA.cmake#L11
-# https://gitlab.kitware.com/cmake/cmake/-/issues/19017
-# For CUDA >= 10.2, we cannot use --compiler-options as '-forward-unknown-to-host-compiler' would be added by default to nvcc options.
-# For the reason, we add "${CMAKE_CXX_FLAGS}" instead of "--compiler-options ${CMAKE_CXX_FLAGS}" here.
-# ==> We changed to use "${CMAKE_CUDA_FLAGS}" instead of "${CMAKE_CXX_FLAGS}" ${CMAKE_CXX_FLAGS} can have wrong options such as '-march=nocona'
-set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -use_fast_math -Xptxas=-v")
-set(CMAKE_CUDA_FLAGS_DEBUG "${CMAKE_CUDA_FLAGS_DEBUG} -G")
-set(CMAKE_CUDA_FLAGS_RELEASE "${CMAKE_CUDA_FLAGS_RELEASE} -lineinfo")
-set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "${CMAKE_CUDA_FLAGS_RELWITHDEBINFO} -lineinfo")
 
 # Include CUDA headers explicitly for VSCode intelli-sense
 include_directories(AFTER SYSTEM ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
@@ -169,11 +157,6 @@ set_target_properties(${CUCIM_PLUGIN_NAME}
         CXX_STANDARD 17
         CXX_STANDARD_REQUIRED YES
         CXX_EXTENSIONS NO
-        CUDA_STANDARD 17
-        CUDA_STANDARD_REQUIRED YES
-        CUDA_EXTENSIONS NO
-        CUDA_SEPARABLE_COMPILATION ON
-        CUDA_RUNTIME_LIBRARY Shared
         SOVERSION ${PROJECT_VERSION_MAJOR}
         VERSION ${PROJECT_VERSION}
 )
diff --git a/cpp/plugins/cucim.kit.cumed/benchmarks/CMakeLists.txt b/cpp/plugins/cucim.kit.cumed/benchmarks/CMakeLists.txt
index 43044a9b5..ade52265f 100644
--- a/cpp/plugins/cucim.kit.cumed/benchmarks/CMakeLists.txt
+++ b/cpp/plugins/cucim.kit.cumed/benchmarks/CMakeLists.txt
@@ -17,18 +17,12 @@
 # Add executable: cumed_benchmarks
 ################################################################################
 add_executable(cumed_benchmarks main.cpp config.h)
-#set_source_files_properties(main.cpp PROPERTIES LANGUAGE CUDA) # failed with CLI11 library
 
 set_target_properties(cumed_benchmarks
     PROPERTIES
         CXX_STANDARD 17
         CXX_STANDARD_REQUIRED YES
         CXX_EXTENSIONS NO
-        CUDA_STANDARD 17
-        CUDA_STANDARD_REQUIRED YES
-        CUDA_EXTENSIONS NO
-        CUDA_SEPARABLE_COMPILATION ON
-        CUDA_RUNTIME_LIBRARY Shared
 )
 target_compile_features(cumed_benchmarks PRIVATE ${CUCIM_REQUIRED_FEATURES})
 # Use generator expression to avoid `nvcc fatal   : Value '-std=c++17' is not defined for option 'Werror'`
diff --git a/cpp/plugins/cucim.kit.cumed/tests/CMakeLists.txt b/cpp/plugins/cucim.kit.cumed/tests/CMakeLists.txt
index 25ae8cffd..95df56068 100644
--- a/cpp/plugins/cucim.kit.cumed/tests/CMakeLists.txt
+++ b/cpp/plugins/cucim.kit.cumed/tests/CMakeLists.txt
@@ -24,18 +24,12 @@ add_executable(cumed_tests
         main.cpp
         test_basic.cpp
         )
-set_source_files_properties(test_basic.cpp PROPERTIES LANGUAGE CUDA) # failed with CLI11 library
 
 set_target_properties(cumed_tests
     PROPERTIES
         CXX_STANDARD 17
         CXX_STANDARD_REQUIRED YES
         CXX_EXTENSIONS NO
-        CUDA_STANDARD 17
-        CUDA_STANDARD_REQUIRED YES
-        CUDA_EXTENSIONS NO
-        CUDA_SEPARABLE_COMPILATION ON
-        CUDA_RUNTIME_LIBRARY Shared
 )
 target_compile_features(cumed_tests PRIVATE ${CUCIM_REQUIRED_FEATURES})
 # Use generator expression to avoid `nvcc fatal   : Value '-std=c++17' is not defined for option 'Werror'`
diff --git a/cpp/plugins/cucim.kit.cuslide/CMakeLists.txt b/cpp/plugins/cucim.kit.cuslide/CMakeLists.txt
index 94e654ace..ce4314677 100644
--- a/cpp/plugins/cucim.kit.cuslide/CMakeLists.txt
+++ b/cpp/plugins/cucim.kit.cuslide/CMakeLists.txt
@@ -28,7 +28,7 @@ file(STRINGS ${CMAKE_CURRENT_LIST_DIR}/VERSION VERSION)
 # Append local cmake module path
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_LIST_DIR}/cmake/modules")
 
-project(cuslide VERSION ${VERSION} DESCRIPTION "cuslide" LANGUAGES C CXX CUDA)
+project(cuslide VERSION ${VERSION} DESCRIPTION "cuslide" LANGUAGES C CXX)
 set(CUCIM_PLUGIN_NAME "cucim.kit.cuslide")
 
 ################################################################################
@@ -72,19 +72,7 @@ endif()
 find_package(CUDAToolkit REQUIRED)
 
 set(CMAKE_CXX_STANDARD 17)
-set(CMAKE_CUDA_STANDARD 17) # Clion issue: https://youtrack.jetbrains.com/issue/CPP-19165 (fixed)
-set(CMAKE_CUDA_STANDARD_REQUIRED YES)
 set(CMAKE_CXX_STANDARD_REQUIRED YES)
-cucim_define_cuda_architectures(60;70;75;80;86)
-# https://github.com/Kitware/CMake/blob/master/Modules/Compiler/NVIDIA-CUDA.cmake#L11
-# https://gitlab.kitware.com/cmake/cmake/-/issues/19017
-# For CUDA >= 10.2, we cannot use --compiler-options as '-forward-unknown-to-host-compiler' would be added by default to nvcc options.
-# For the reason, we add "${CMAKE_CXX_FLAGS}" instead of "--compiler-options ${CMAKE_CXX_FLAGS}" here.
-# ==> We changed to use "${CMAKE_CUDA_FLAGS}" instead of "${CMAKE_CXX_FLAGS}" ${CMAKE_CXX_FLAGS} can have wrong options such as '-march=nocona'
-set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -use_fast_math -Xptxas=-v")
-set(CMAKE_CUDA_FLAGS_DEBUG "${CMAKE_CUDA_FLAGS_DEBUG} -G")
-set(CMAKE_CUDA_FLAGS_RELEASE "${CMAKE_CUDA_FLAGS_RELEASE} -lineinfo")
-set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "${CMAKE_CUDA_FLAGS_RELWITHDEBINFO} -lineinfo")
 
 # Include CUDA headers explicitly for VSCode intelli-sense
 include_directories(AFTER SYSTEM ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
@@ -123,7 +111,6 @@ add_definitions(-D_GLIBCXX_USE_CXX11_ABI=0) # TODO: create two library, one with
 ################################################################################
 # Define dependencies
 ################################################################################
-#superbuild_depend(cucim)
 superbuild_depend(fmt)
 superbuild_depend(libjpeg-turbo) # libjpeg-turbo should be located before libtiff as libtiff depends on libjpeg-turbo
 superbuild_depend(libopenjpeg)
@@ -211,11 +198,6 @@ set_target_properties(${CUCIM_PLUGIN_NAME}
         CXX_STANDARD 17
         CXX_STANDARD_REQUIRED YES
         CXX_EXTENSIONS NO
-        CUDA_STANDARD 17
-        CUDA_STANDARD_REQUIRED YES
-        CUDA_EXTENSIONS NO
-        CUDA_SEPARABLE_COMPILATION ON
-        CUDA_RUNTIME_LIBRARY Shared
         SOVERSION ${PROJECT_VERSION_MAJOR}
         VERSION ${PROJECT_VERSION}
 )
@@ -228,7 +210,6 @@ target_link_libraries(${CUCIM_PLUGIN_NAME}
         PRIVATE
             CUDA::cudart
             deps::fmt
-            #cucim::rmm
             cucim::cucim
             deps::libtiff
             deps::libjpeg-turbo
diff --git a/cpp/plugins/cucim.kit.cuslide/benchmarks/CMakeLists.txt b/cpp/plugins/cucim.kit.cuslide/benchmarks/CMakeLists.txt
index d0a405c37..490860c39 100644
--- a/cpp/plugins/cucim.kit.cuslide/benchmarks/CMakeLists.txt
+++ b/cpp/plugins/cucim.kit.cuslide/benchmarks/CMakeLists.txt
@@ -24,11 +24,6 @@ set_target_properties(cuslide_benchmarks
         CXX_STANDARD 17
         CXX_STANDARD_REQUIRED YES
         CXX_EXTENSIONS NO
-        CUDA_STANDARD 17
-        CUDA_STANDARD_REQUIRED YES
-        CUDA_EXTENSIONS NO
-        CUDA_SEPARABLE_COMPILATION ON
-        CUDA_RUNTIME_LIBRARY Shared
 )
 target_compile_features(cuslide_benchmarks PRIVATE ${CUCIM_REQUIRED_FEATURES})
 # Use generator expression to avoid `nvcc fatal   : Value '-std=c++17' is not defined for option 'Werror'`
diff --git a/cpp/plugins/cucim.kit.cuslide/tests/CMakeLists.txt b/cpp/plugins/cucim.kit.cuslide/tests/CMakeLists.txt
index e998219b2..c779d63b8 100644
--- a/cpp/plugins/cucim.kit.cuslide/tests/CMakeLists.txt
+++ b/cpp/plugins/cucim.kit.cuslide/tests/CMakeLists.txt
@@ -26,18 +26,11 @@ add_executable(cuslide_tests
         test_read_rawtiff.cpp
         test_philips_tiff.cpp
         )
-set_source_files_properties(test_read_rawtiff.cpp PROPERTIES LANGUAGE CUDA) # failed with CLI11 library
-
 set_target_properties(cuslide_tests
     PROPERTIES
         CXX_STANDARD 17
         CXX_STANDARD_REQUIRED YES
         CXX_EXTENSIONS NO
-        CUDA_STANDARD 17
-        CUDA_STANDARD_REQUIRED YES
-        CUDA_EXTENSIONS NO
-        CUDA_SEPARABLE_COMPILATION ON
-        CUDA_RUNTIME_LIBRARY Shared
 )
 target_compile_features(cuslide_tests PRIVATE ${CUCIM_REQUIRED_FEATURES})
 # Use generator expression to avoid `nvcc fatal   : Value '-std=c++17' is not defined for option 'Werror'`
@@ -52,6 +45,7 @@ target_compile_definitions(cuslide_tests
 )
 target_link_libraries(cuslide_tests
         PRIVATE
+            CUDA::cudart
             cucim::cucim
             ${CUCIM_PLUGIN_NAME}
             deps::catch2
diff --git a/cpp/plugins/cucim.kit.cuslide/tests/test_read_rawtiff.cpp b/cpp/plugins/cucim.kit.cuslide/tests/test_read_rawtiff.cpp
index f76b83db6..54f4801ac 100644
--- a/cpp/plugins/cucim.kit.cuslide/tests/test_read_rawtiff.cpp
+++ b/cpp/plugins/cucim.kit.cuslide/tests/test_read_rawtiff.cpp
@@ -59,7 +59,7 @@
 static void shuffle_offsets(uint32_t count, uint64_t* offsets, uint64_t* bytecounts)
 {
     // Fisher-Yates shuffle
-    for (int i = 0; i < count; ++i)
+    for (uint32_t i = 0; i < count; ++i)
     {
         int j = (std::rand() % (count - i)) + i;
         std::swap(offsets[i], offsets[j]);
@@ -172,7 +172,7 @@ TEST_CASE("Verify raw tiff read", "[test_read_rawtiff.cpp]")
             {
                 cucim::logger::Timer timer("- read whole : {:.7f}\n", true, false);
 
-                ssize_t read_cnt = fd->pread(aligned_host, test_file_size, 0);
+                fd->pread(aligned_host, test_file_size, 0);
 
                 double elapsed_time = timer.stop();
                 if (iter >= skip_count)
@@ -194,7 +194,7 @@ TEST_CASE("Verify raw tiff read", "[test_read_rawtiff.cpp]")
 
                 for (uint32_t i = 0; i < image_piece_count_; ++i)
                 {
-                    ssize_t read_cnt = fd->pread(aligned_host, image_piece_bytecounts_[i], image_piece_offsets_[i]);
+                    fd->pread(aligned_host, image_piece_bytecounts_[i], image_piece_offsets_[i]);
                 }
 
                 double elapsed_time = timer.stop();
@@ -220,7 +220,7 @@ TEST_CASE("Verify raw tiff read", "[test_read_rawtiff.cpp]")
             {
                 cucim::logger::Timer timer("- read whole : {:.7f}\n", true, false);
 
-                ssize_t read_cnt = fd->pread(aligned_host, test_file_size, 0);
+                fd->pread(aligned_host, test_file_size, 0);
 
                 double elapsed_time = timer.stop();
                 if (iter >= skip_count)
@@ -242,7 +242,7 @@ TEST_CASE("Verify raw tiff read", "[test_read_rawtiff.cpp]")
 
                 for (uint32_t i = 0; i < image_piece_count_; ++i)
                 {
-                    ssize_t read_cnt = fd->pread(buffer_host, image_piece_bytecounts_[i], image_piece_offsets_[i]);
+                    fd->pread(buffer_host, image_piece_bytecounts_[i], image_piece_offsets_[i]);
                 }
 
                 double elapsed_time = timer.stop();
@@ -274,7 +274,7 @@ TEST_CASE("Verify raw tiff read", "[test_read_rawtiff.cpp]")
             {
                 cucim::logger::Timer timer("- preload : {:.7f}\n", true, false);
 
-                ssize_t read_cnt = fd->pread(aligned_host, large_block_size, file_start_offset);
+                fd->pread(aligned_host, large_block_size, file_start_offset);
 
                 double elapsed_time = timer.stop();
                 if (iter >= skip_count)
@@ -358,7 +358,7 @@ TEST_CASE("Verify raw tiff read", "[test_read_rawtiff.cpp]")
                 for (uint32_t i = 0; i < image_piece_count_; ++i)
                 {
                     // 3.441 => 3.489
-                    ssize_t read_cnt = fd->pread(buffer_host, image_piece_bytecounts_[i], image_piece_offsets_[i]);
+                    fd->pread(buffer_host, image_piece_bytecounts_[i], image_piece_offsets_[i]);
                     //                                        memcpy(buffer_host, static_cast<char*>(mmap_host) +
                     //                                        image_piece_offsets_[i], image_piece_bytecounts_[i]);
                 }
diff --git a/cpp/src/memory/memory_manager.cu b/cpp/src/memory/memory_manager.cpp
similarity index 86%
rename from cpp/src/memory/memory_manager.cu
rename to cpp/src/memory/memory_manager.cpp
index 54ad52c1d..cb77d2bd8 100644
--- a/cpp/src/memory/memory_manager.cu
+++ b/cpp/src/memory/memory_manager.cpp
@@ -14,6 +14,8 @@
  * limitations under the License.
  */
 
+#define CUCIM_EXPORTS // For exporting functions globally
+
 #include "cucim/memory/memory_manager.h"
 
 #include <memory_resource>
@@ -25,7 +27,6 @@
 #include "cucim/profiler/nvtx3.h"
 #include "cucim/util/cuda.h"
 
-
 CUCIM_API void* cucim_malloc(size_t size)
 {
     PROF_SCOPED_RANGE(PROF_EVENT_P(cucim_malloc, size));
@@ -71,13 +72,13 @@ void get_pointer_attributes(PointerAttributes& attr, const void* ptr)
     }
 }
 
-bool move_raster_from_host(void** target, size_t size, const cucim::io::Device& dst_device)
+CUCIM_API bool move_raster_from_host(void** target, size_t size, const cucim::io::Device& dst_device)
 {
     switch (dst_device.type())
     {
     case cucim::io::DeviceType::kCPU:
         break;
-    case cucim::io::DeviceType::kCUDA:
+    case cucim::io::DeviceType::kCUDA: {
         cudaError_t cuda_status;
         void* host_mem = *target;
         void* cuda_mem;
@@ -93,11 +94,15 @@ bool move_raster_from_host(void** target, size_t size, const cucim::io::Device&
         }
         cucim_free(host_mem);
         *target = cuda_mem;
+        break;
+    }
+    default:
+        throw std::runtime_error("Unsupported device type");
     }
     return true;
 }
 
-bool move_raster_from_device(void** target, size_t size, const cucim::io::Device& dst_device)
+CUCIM_API bool move_raster_from_device(void** target, size_t size, const cucim::io::Device& dst_device)
 {
     switch (dst_device.type())
     {
@@ -116,6 +121,8 @@ bool move_raster_from_device(void** target, size_t size, const cucim::io::Device
     }
     case cucim::io::DeviceType::kCUDA:
         break;
+    default:
+        throw std::runtime_error("Unsupported device type");
     }
     return true;
 }
diff --git a/cpp/tests/CMakeLists.txt b/cpp/tests/CMakeLists.txt
index 32b082bdd..eb5910e33 100644
--- a/cpp/tests/CMakeLists.txt
+++ b/cpp/tests/CMakeLists.txt
@@ -25,18 +25,12 @@ add_executable(cucim_tests
         test_cufile.cpp
         test_metadata.cpp
         )
-set_source_files_properties(main.cpp test_read_region.cpp test_cufile.cpp test_metadata.cpp PROPERTIES LANGUAGE CUDA)
 
 set_target_properties(cucim_tests
     PROPERTIES
         CXX_STANDARD 17
         CXX_STANDARD_REQUIRED YES
         CXX_EXTENSIONS NO
-        CUDA_STANDARD 17
-        CUDA_STANDARD_REQUIRED YES
-        CUDA_EXTENSIONS NO
-        CUDA_SEPARABLE_COMPILATION ON
-        CUDA_RUNTIME_LIBRARY Shared
 )
 target_compile_features(cucim_tests PRIVATE ${CUCIM_REQUIRED_FEATURES})
 # Use generator expression to avoid `nvcc fatal   : Value '-std=c++17' is not defined for option 'Werror'`
@@ -51,6 +45,7 @@ target_compile_definitions(cucim_tests
 )
 target_link_libraries(cucim_tests
         PRIVATE
+            CUDA::cudart
             ${CUCIM_PACKAGE_NAME}
             deps::catch2
             deps::openslide
diff --git a/cpp/tests/test_cufile.cpp b/cpp/tests/test_cufile.cpp
index be037fcff..a1330d52b 100644
--- a/cpp/tests/test_cufile.cpp
+++ b/cpp/tests/test_cufile.cpp
@@ -64,6 +64,7 @@ static void create_test_file(const char* file_name, int size)
         test_data[i] = ::rand() % 256; // or i % 256;
     }
     ssize_t write_cnt = write(fd, test_data, size);
+    (void)write_cnt;
     assert(write_cnt == size);
     close(fd);
 }
@@ -125,8 +126,8 @@ TEST_CASE("Verify libcufile usage", "[test_cufile.cpp]")
                 INFO("# unaligned_host");
                 size_t reference_whole_hash = 0;
                 size_t reference_hash = 0;
-                size_t reference_write_cnt = 0;
-                size_t reference_read_cnt = 0;
+                ssize_t reference_write_cnt = 0;
+                ssize_t reference_read_cnt = 0;
                 for (int flag_idx = 0; flag_idx < W_FLAG_LEN; ++flag_idx)
                 {
                     INFO(fmt::format("flag_index: {} ({})\n  count: {}\n  file_offset: {}  buf_offset: {}\n", flag_idx,
@@ -185,8 +186,8 @@ TEST_CASE("Verify libcufile usage", "[test_cufile.cpp]")
                 INFO("# aligned_host");
                 size_t reference_whole_hash = 0;
                 size_t reference_hash = 0;
-                size_t reference_write_cnt = 0;
-                size_t reference_read_cnt = 0;
+                ssize_t reference_write_cnt = 0;
+                ssize_t reference_read_cnt = 0;
                 for (int flag_idx = 0; flag_idx < W_FLAG_LEN; ++flag_idx)
                 {
                     INFO(fmt::format("flag_index: {} ({})\n  count: {}\n  file_offset: {}  buf_offset: {}\n", flag_idx,
@@ -245,8 +246,8 @@ TEST_CASE("Verify libcufile usage", "[test_cufile.cpp]")
                     INFO("# unaligned_device");
                     size_t reference_whole_hash = 0;
                     size_t reference_hash = 0;
-                    size_t reference_write_cnt = 0;
-                    size_t reference_read_cnt = 0;
+                    ssize_t reference_write_cnt = 0;
+                    ssize_t reference_read_cnt = 0;
                     for (int flag_idx = 0; flag_idx < W_FLAG_LEN; ++flag_idx)
                     {
                         INFO(fmt::format("flag_index: {} ({})\n  count: {}\n  file_offset: {}  buf_offset: {}\n", flag_idx,
@@ -306,8 +307,8 @@ TEST_CASE("Verify libcufile usage", "[test_cufile.cpp]")
                     INFO("# aligned_device");
                     size_t reference_whole_hash = 0;
                     size_t reference_hash = 0;
-                    size_t reference_write_cnt = 0;
-                    size_t reference_read_cnt = 0;
+                    ssize_t reference_write_cnt = 0;
+                    ssize_t reference_read_cnt = 0;
                     for (int flag_idx = 0; flag_idx < W_FLAG_LEN; ++flag_idx)
                     {
                         INFO(fmt::format("flag_index: {} ({})\n  count: {}\n  file_offset: {}  buf_offset: {}\n", flag_idx,
@@ -368,8 +369,8 @@ TEST_CASE("Verify libcufile usage", "[test_cufile.cpp]")
                     INFO("# unaligned_device (pinned host)");
                     size_t reference_whole_hash = 0;
                     size_t reference_hash = 0;
-                    size_t reference_write_cnt = 0;
-                    size_t reference_read_cnt = 0;
+                    ssize_t reference_write_cnt = 0;
+                    ssize_t reference_read_cnt = 0;
                     for (int flag_idx = 0; flag_idx < W_FLAG_LEN; ++flag_idx)
                     {
                         INFO(fmt::format("flag_index: {} ({})\n  count: {}\n  file_offset: {}  buf_offset: {}\n", flag_idx,
@@ -431,8 +432,8 @@ TEST_CASE("Verify libcufile usage", "[test_cufile.cpp]")
                     INFO("# aligned_device (pinned host)");
                     size_t reference_whole_hash = 0;
                     size_t reference_hash = 0;
-                    size_t reference_write_cnt = 0;
-                    size_t reference_read_cnt = 0;
+                    ssize_t reference_write_cnt = 0;
+                    ssize_t reference_read_cnt = 0;
                     for (int flag_idx = 0; flag_idx < W_FLAG_LEN; ++flag_idx)
                     {
                         INFO(fmt::format("flag_index: {} ({})\n  count: {}\n  file_offset: {}  buf_offset: {}\n", flag_idx,
@@ -496,8 +497,8 @@ TEST_CASE("Verify libcufile usage", "[test_cufile.cpp]")
                     INFO("# unaligned_device (managed)");
                     size_t reference_whole_hash = 0;
                     size_t reference_hash = 0;
-                    size_t reference_write_cnt = 0;
-                    size_t reference_read_cnt = 0;
+                    ssize_t reference_write_cnt = 0;
+                    ssize_t reference_read_cnt = 0;
                     for (int flag_idx = 0; flag_idx < W_FLAG_LEN; ++flag_idx)
                     {
                         INFO(fmt::format("flag_index: {} ({})\n  count: {}\n  file_offset: {}  buf_offset: {}\n", flag_idx,
@@ -560,8 +561,8 @@ TEST_CASE("Verify libcufile usage", "[test_cufile.cpp]")
                     INFO("# aligned_device (managed)");
                     size_t reference_whole_hash = 0;
                     size_t reference_hash = 0;
-                    size_t reference_write_cnt = 0;
-                    size_t reference_read_cnt = 0;
+                    ssize_t reference_write_cnt = 0;
+                    ssize_t reference_read_cnt = 0;
                     for (int flag_idx = 0; flag_idx < W_FLAG_LEN; ++flag_idx)
                     {
                         INFO(fmt::format("flag_index: {} ({})\n  count: {}\n  file_offset: {}  buf_offset: {}\n", flag_idx,
diff --git a/examples/cpp/CMakeLists.txt b/examples/cpp/CMakeLists.txt
index 827921447..9988e985f 100644
--- a/examples/cpp/CMakeLists.txt
+++ b/examples/cpp/CMakeLists.txt
@@ -18,18 +18,12 @@
 ################################################################################
 
 add_executable(tiff_image tiff_image/main.cpp)
-set_source_files_properties(tiff_image/main.cpp PROPERTIES LANGUAGE CUDA)
 
 set_target_properties(tiff_image
     PROPERTIES
         CXX_STANDARD 17
         CXX_STANDARD_REQUIRED YES
         CXX_EXTENSIONS NO
-        CUDA_STANDARD 17
-        CUDA_STANDARD_REQUIRED YES
-        CUDA_EXTENSIONS NO
-        CUDA_SEPARABLE_COMPILATION ON
-        CUDA_RUNTIME_LIBRARY Shared
 )
 target_compile_features(tiff_image PRIVATE ${CUCIM_REQUIRED_FEATURES})
 # Use generator expression to avoid `nvcc fatal   : Value '-std=c++17' is not defined for option 'Werror'`
diff --git a/gds/CMakeLists.txt b/gds/CMakeLists.txt
index b8a959840..019f9f891 100644
--- a/gds/CMakeLists.txt
+++ b/gds/CMakeLists.txt
@@ -32,27 +32,18 @@ add_library(cufile_stub
         src/cufile_stub.cpp
         )
 
-set_source_files_properties(src/cufile_stub.cpp PROPERTIES LANGUAGE CUDA)
-
 # Compile options
 set_target_properties(cufile_stub
     PROPERTIES
         CXX_STANDARD 17
         CXX_STANDARD_REQUIRED YES
         CXX_EXTENSIONS NO
-        CUDA_STANDARD 17
-        CUDA_STANDARD_REQUIRED YES
-        CUDA_EXTENSIONS NO
-        CUDA_SEPARABLE_COMPILATION ON
-        CUDA_RUNTIME_LIBRARY Shared
         # To prevent the error message: /usr/bin/ld: ../lib/libcufile_stub.a(cufile_stub.cpp.o): relocation R_X86_64_PC32 against symbol `__fatbinwrap_46_tmpxft_00005869_00000000_6_cufile_stub_cpp1_ii_1e2ddd49' can not be used when making a shared object; recompile with -fPIC
         POSITION_INDEPENDENT_CODE ON
-#        SOVERSION ${PROJECT_VERSION_MAJOR}
-#        VERSION ${PROJECT_VERSION}
 )
 
 # Note: Looks like the following line causes error on CMake 3.18.4 (it is working on 3.18.2). Keeping it for now.
-set(CUCIM_REQUIRED_FEATURES cxx_std_17 cuda_std_17)
+set(CUCIM_REQUIRED_FEATURES cxx_std_17)
 target_compile_features(cufile_stub PRIVATE ${CUCIM_REQUIRED_FEATURES})
 # Use generator expression to avoid `nvcc fatal   : Value '-std=c++17' is not defined for option 'Werror'`
 target_compile_options(cufile_stub
@@ -64,6 +55,8 @@ target_compile_options(cufile_stub
 target_link_libraries(cufile_stub
         PUBLIC
             ${CMAKE_DL_LIBS}
+        PRIVATE
+            CUDA::cudart
         )
 
 # Set GDS include path (cufile.h)