How to apply this in case of a CNN model?

[en-ashay]

For parametric test, can I directly provide the model accuracy and test data size. Also since I am providing the mean value as the accuracy obtained ( avg of all on test data) . What is the assumption here:

1. The accuracy value of all individual values (images) will form a normal distribution
2. The final accuracy of model when tested multiple times will follow a normal distribution

Does we make 1st or 2nd assumption or both for parametric case?