Keras Custom Batch Normalization layer with an extra variable that can be changed in run time

Question

I have implemented a custom version of Batch Normalization with adding self.skip variable that act somehow as trainable. Here is the minimal code:<