What are the best practices for writing a custom Keras Layer in regard to reduced duplicate code, configuration, and storage?

Question

I am a bit confused about writing a Keras Layer following the best practices. It feels like there is a bit of duplicate code and duplicate configuration.

Specifically, m