So, we have 3 recent polls, using 3 different criterias:
NBC poll: Clinton+10 (among all adults)
PPP poll: Clinton+5 (among RV)
Rasmussen poll: Clinton+2 (among LV)
It's far too early to poll likely voters at this stage. That can be done two months in advance, not Three years.
And both parties have three more years to register more voters, especially in the important battleground states. Also, an awful lot of people are going turn 18 during the next three years and almost as many older voters are likely to die off. Thus I consider the all voters model the most credible at this stage, followed by the registered voters. I think the all voters model is the best one until 6-9 months prior to the election. And from two months on, it should be likely voters.