I'm pretty sure Trump won white college educated voters based on everything I've seen from precinct data. It's much more likely that exit polls overestimated his support with minorities, as they did with Bush. The gap between the polls and the actual results is a classic example of social desirability bias. Voting for Trump was politically incorrect and therefore stigmatized in the suburbs in a way that it couldn't be stigmatized in rural areas where support for him was overwhelming.
This phenomenon has been seen before in cases such as elections between a white and a nonwhite candidate (the Bradley effect) and on the issue of gay marriage (
http://journals.sagepub.com/doi/pdf/10.1177/1532673X13484791). This phenomenon may have also played a role in pollsters getting Brexit wrong, as well as the fact that "far-right" parties across Europe almost always outperform polls.