What elm-bytes is good at is decoding binary serialized data. So it seems possible to (automatically) write capnproto decoders using elm-bytes.
But because elm is a more high-level language, the performance characteristics will be different. In elm, you cannot put some bytes in a piece of memory and say “you are of type X now” like you could in C, so the in-memory representation will be different and there is a cost to going back and forth between bytes and actual elm data structures.
But I think this would be an interesting project. This kind of tooling really helps when you have a lot of decoders to write, and it can do optimizations that would just be too tedious if done by hand. The performance will be reasonable (though that does depend on the kind of data you send).
So I’d say see if you can make a prototype that feels nice to work with, and then it can be optimized.