Yes you can do that and shard a very large model across the devices but it's way too slow so you will get no performance gains beyond being able to run a much larger model at all.