I'd be going back to asking why you want to split it off though. If there's legitimate reasons for that then my other concern would be what about areas of the existing codebase that are intertwined with those messages' modelling. Let's say you use the messages as objects throughout the system and those objects have their own methods that give you some nice way to intepret / action those messages. Now what? Do you need to re-write all that, create a shared library of models etc? It does add complexity and I'd always be last-resorting it.