Jay has good points. If the course can be ridden in a repeatable fashion, and the turns can be easily identified and used, then this is a good situation for a certification.
The question still remains, though - at what point (absolute distance, or percentage of course) do we need to be calibrating on the course's predominant surface, and not pavement? I know courses will be long if calibrated on pavement then measured on grass and gravel, but aren't we still interested in accuracy? If a 5k X-C course tapes-out to 5250 meters, it is "at least 5k". But, it's 5% long. Are we good with that?