Last things first...
Human feel: get human by being human, complete with all of our lil' imperfections. Er, um, sorry, I was getting a little too philosophical there! But seriously though, if you want human, play in your parts and don't quantize them.
As you start to pile up parts in an orchestration there may reach a point where one part now sounds a little too out of time with the others. The only practical solution (IMO) is to re-play the part.
Next, to control the emotion of the part while you play it in (crescendo, decrescendo, etc.) you also have to develop technique for playing in your parts while you ride the controller (modwheel or whatever you're using to control dynamics). But the way you've probably got things set up now, you have not only CC control of dynamics but also velocity control of dynamics. Your choices are to either remove all velocity-induced dynamics from the sound and control levels exclusively with your CC, or, try to use both. Which is the best way to go is hard to say. It all depends on the nature of the part you're playing.
"Expression" is just a name given to CC#11. It's nothing more than a volume control, just like CC#7. In some plugins, CC#11 ("expression") controls both volume and filter, just like I suggested you configure in the mod matrix. However, you can't use it to control volume/filter in EXS-24 because CC#11 is hard-wired in EXS to control the volume of the plugin itself. The difference between CC#11 and CC#7 is this: CC#7 will move the instrument channel's fader, while CC#11 will adjust the volume of the instrument itself without any visual feedback (other than whatever you see in the hyperdraw display).
Will post back later with some more thoughts.