If you mean a voice with three layers, I think you need to use MIDI for that.
You can create all sorts of splits and layers by sending the keyboard's Note events to a DAW and then back to the keyboard. The way I do it is to send channel 1 to the DAW, then send the data back to the keyboard on other channels, filtering each channel's data-- and possibly transposing it, as well-- based on a range of notes. The details of where and how this is done will depend on the DAW.
You may or may not be able to do this without having to go through a DAW, because XG includes Multi-Part messages which might be able to accomplish this. Joe H can probably tell you whether it can be done, and if so, how to do it.