I want to make custom model in tensorflow. I made functions to create basic layers, like Conv2D, Dense, Flatten. I stuck with batch normalization implementation.