Deploying multiple Comprehend Custom Classifiers (multi-label mode)

I want to train and deploy multiple comprehend custom classifiers (for example 50 models). I want to be able to classify my documents in near real-time (a couple of seconds are fine) 24/7. The problem is that deploying one end-point for each classifier is very expensive, especially that one or two IU would be enough for all my models combined (I am expecting to process around 10 document a minute total/length of one document is around 1000 characters ). Is there a way where I can deploy multiple models behind the same endpoint (similar to the multi-model endpoint in SageMaker)? Or maybe do an asynchronous approach and somewho make sure I get the response within seconds?