It wasn't mentioned what you are upgrading from/to. There are several methods and generally the cleanest way is probably just to do an in-place head swap. You can also as mentioned add the nodes temporarily to the cluster, or even hook up a second cluster and SnapMirror everything over.
If you're working with your account team on this, generally they include PS hours to do the install for you, or they may have guidance for your specific workloads. I'd suggest working with them on the best methods.
