By the time you get to the rigging stage you should have all the modeling portion signed off anyway (in a production environment) If you have one mesh to rig, for example a human which is all one piece of geo, then I would say just go ahead and do a normal bind. If the model is broken up into sections and is extremely complex, like for example a robot or similar. Then I would group the sections (like right leg, left leg etc) together and do a parent constrain to the joints, this will simply make it easier for searching through a hypergraph.