Unfortunately I do not know any e-book about the subject. I try to summarize my experiences with this kind of designs, without actually being any guru in this particular topics -- but I have made several basically similar constructions in different contexts. Some well working, others needing rework.
I would follow these principles: You may keep the power output stage, inclusive its bypass capacitors, in one ground system/feed system. The two channels are likely not disturbing each others, because they have relatively high level inputs. The power output stage is fed directly from the power supply.
When the load is not ground referenced, the return currents travel through power rails. In any case thus the bypass/filter caps are very important. Be aware the ESR values of those cpacitors, too! The Effective Series Resistance of electrolytics is often the most important factor for the ripple and noise distribution. It is easy to calculate the minimum capacitance, and then be surprised that the result is not even close by matching the theory.
The driver driving that output stage has intermediate signal levels and moderate return currents. Do not let the ouput return currents run through this stage, and a separate filtering might be necessary--I am not sure about your particular design. Input of this is connected to the most sensitive input part, and ground to it should be connected from here in single point, parallel with the signal(s).
The input amplifier should have at least a RC filtered power supply feed, or even it's own regulator chips. And of course, here the "single-point-grounding" FOR THE INPUT AMPLIFER itself should be then carefully connected to the rest, to avoid the "ground loops" and also unwanted feedback from the output stage.
Try to imagine where the big currents flow in the output, and keep those paths short, low-impedance and separate. Similar way, the sensitive parts should be an "islands" of their own, with no ripple or output currents flowing through them.
Good luck!
Ted