Model compression and simplification pipelines for fast deep neural network inference in FPGAs in HEP