Improve device handling

[amacati]

We are currently threading devices through all struct inits. We could potentially reduce that to a single device placement after creating the data, which removes a lot of code and simplifies sharding