While considering the tradeoffs between different DDS vendors, and different middleware altogether, I started wondering how much room there would be to customize your system's use of different middleware's for different purposes.

I know you can set QoS settings at the publisher/subscriber/server/client level, and you can choose middleware at runtime... but could you choose middleware at the publisher/subscriber/server/client level?

Would it, for instance, be possible to use one middleware implementation for a robot's (onboard) inter/intra process communication where latency and bandwidth are less of a concern, and select a different middleware implementation for communication between different machines (e.g between a robot and a ground node with high latency or low reliability)?

If possible, what would the drawbacks be? Presumably you'll suffer a hit on the memory footprint by loading multiple middleware libraries?

I'm definitely not an expert in networking or middleware so any feedback to satisfy my curiosity would be great. Cheers!