How to deploy/use externally trained custom tensorflow model on deeplens?

Question

I have trained a custom image captioning model with my own model classes using the TensorFlow framework, I want to deploy it on AWS Deeplens but the problem is how to convert the classes to the metadata and create the artifacts in a proper way and save it to s3, so that I can deploy it. Can't find any resource regarding this. If anyone has already done it please share the process. Thanks