Hi Kedu,
Regarding my suggestion that it is OK, or even a bit easier/better, to specify perpendicularity on each of the two planar faces that are used to establish datum B, instead of placing the perpendicularity under the width dimension to control the center plane:
1) Establishing a datum reference frame is (or at least should be) separate and independent from the process of qualifying the datum features. As long as the datum features (both planar surfaces) have all controllable characteristics relative to datum A controlled, so their form and their orientation WRT datum A, then all is completely fine. An example exactly like this case does not need to be shown in Y14.5. What I am recommending is fully supported by the concepts in the standard.
2) Datum B is the center plane, but it were done correctly, it is a center plane that is perfectly perpendicular to datum A. It's a different center plane, the center plane of the feature, which is the center plane of an unrelated actual mating envelope, that a perpendicularity tolerance placed under the width dimension would apply to. That feature center plane would be bounded by the diameter of the cylindrical pockets so the requirement has a level or two of complexity that a perpendicularity that is applied directly to each surface would not have. Especially if the width feature that is included in PC-DMIS (which I believe will fit an unrelated mating envelope to the data, from which a center plane could be found) isn't available to the metrologist, most measurements for perpendicularity of a center plane will be evaluated as perpendicularity of the derived median plane. We know Rwelch9 isn't using PC-DMIS yet, so why recommend a tolerance that his CMM software won't evaluate correctly?
3) When we look at the two planar surfaces from a top view (looking down through the part, normal to datum A) the perpendicularity on each of the two planar surfaces does nothing to control whether they are parallel to each other, but that issue is the same for perpendicularity applied the center plane as well. To control the parallelism of the two planar surfaces as viewed from the top, we're relying on the range of the size (width) tolerance. If the whole part looks like a parallelogram when viewed from above, then this may (emphasis on may) get exposed by a position tolerance on the two cylindrical pockets WRT [A,B]. Now I'll bring up again that I like the two cylindrical pockets better as datum feature B. If that change were made then then each of the planar surfaces could have |perp|x.xx|A|B| applied, and their parallelism to each other would be controlled.
Hi Rwelch9,
If the pockets are so shallow that you can probe at only one depth, so a circle is all you can get instead of a cylinder, then your situation is as if you're dealing with holes punched in sheetmetal. With only a center point from each of the two cylinders (circles) you can't measure coaxiality. You can measure the location of each center point relative to the datum reference frame [A,B], with A and B as shown in your sketch, but that's the best you can do. I would not recommend using any other external features of the part as datum features (as I think I saw recommended in past comments) for this measurement, since they're not functional. Data relative to a non-functional datum reference frame is... Non-functional. The A and B you show in your sketch appear to be the best that you have to evaluate the location of the two pockets. That said, I think it would be better overall to switch B and C, as I mentioned in item #3 above.
Regarding a good GD&T book, others here may have a better answer for you. When I read someone else's GD&T book, I look at if from the point of view of trying to make my book better than theirs

. As I sit here this morning, my next task is to finish a revision of the book I use when I provide training. I'm converting from something that was limited to me filling in a bunch of missing information with a tablet PC as we go through the course, to a stand-alone book, with all that info included. I'm quite biased of course, so I'll give a general answer that Amazon has some GD&T books listed (mine not yet included there). I hope someone else will offer some suggestions.
Dean