Tensorflow might not be the fastest in terms of computation speed, but it can be used from research to production with Tensorflow Serving.

As such you won't need to implement/convert your model in another format for usage.