I find it helpful to think of a datum as a feature of a fixture, not a feature of the part.
So, theoretically, you could machine a fixture to have a surface perfectly complementary to a perfectly contoured blade, rest the part on the fixture, and measure deviations over the entire surface.
For a physical fixture, that is a practically impossible thing to do, at least in part because it can only cover the case where the part surface lies entirely to one side of the theoretical surface.
For a CMM, I think it's less impossible, or slightly less impractical, but I think you need to call the theoretical surface a quaternary datum (if that's the right GD-speak), and use some features of the blade root interface as the primary and secondary and tertiary datums, so the inspector just has to bolt the blade to a simple fixture, and then map the blade's surfaces relative to those blade root datums and the theoretical surface.
... as opposed to first mapping the blade's surface and trying to then derive where the datum might best fit on or in it in such a way that the blade root features stay within tolerance, which is sort of what I'm reading into your problem statement.
Pembroke Pines, FL, USA