We have developed a simple software to show how a custom keras model can be automatically translated into c-code. The generated c-code can, in combination with the ARM-CMSIS-NN functions, be used to run neural-net calculations in an efficient way on an embedded micro-controller such as the CORTEX-M4.
The example software on GitHub has also firmware which runs on the STM32F4-Discorevy Board. Part of the firmware was generated with cubeMX.
The example software has a MNIST classifier which can classify handwritten digits.
See https://github.com/InES-HPMM/k2arm for more Details.