My approach with VSL differs depending on the type of piece I'm composing. I'll outline the main approaches to each below:
For classical compositions intended for human performance, I compose the music in Finale (I stick with the Garritan sounds in Finale), add articulation/dyanmics/tempi, markings as needed, prior to exporting the midi files to Cubase. I import the file in to to my Cubase template which is organized by orchestral section, followed by tracks for synths, and audio (live instruments/vocals, etc.) At this point, I'll decide whether or not to actually perform the individual instrumental tracks using the piano, or just add articulations to the midi tracks. If I perform one or more parts, I'll first duplicate the track and retain the midi file, then play in the part using simply a sustain keyswitch for the whole thing. This way, depending on the complexity of the part, I can use a combination of the midi notes from Finale and my live playing (often, the slower, more lyrical sections beneft from my live performance).
Once the parts are in, I play each part back at a slower tempo, and record the keyswitches in. I have set up the lowest octave of the keyboard to accommodate keyswitches from C1-B1, with similar articulations and variations occupying parts of each matrix. I have ensured (as per Beat Kaufmann's suggestion) that the articulations assigned to various keys is as close as possible throughout the range of orchestral instruments (save for the occasions where certain instruments don't have a specific articulation, at which point, I've got alternate ones assigned to those keys). This system has worked well for me, but I only have the standard libraries, and have often wondered how I'll adapt things with the additional articulations if I ever am fortunate enough to acquire the full libraries.
I should also mention, my wife, Becky, has created many combination articulations (such as crossfading a szforzando with the legato patch) for more control beyond what the standard articulations offer.
Now, if I'm composing in the daw, such as my recent posting about my cinematic orchestral album, I'll compose everything using a single sustain keyswitch (but still have my template open with all available articulations). I find that as a performer, I can often create a fairly convincing performance with simply that one keyswitch, as it seems to "read my mind" and react to the various velocities of a live performance better than the imported midi notes would, if they were played back using only a sustain keyswitch. I know this is silly, because at the end of the day, note on/note off/velocity are all simply a numeric reference in midi, and it doesn't have the ability to make my live playing any more musical than midi data, but it "seems" like my pieces are more alive when I play them in myself (of course, there are numerous other things like tempo, note length, etc. that a live performance add to the mix).
But I digress... after the piece has been composed via the one sustain patch, I'll play the track back and record my keyswitches (again playing the track back at a slow enough tempo to record my keyswitches in one take). After that's all done, I'll refine and change articulations as needed. All of this, of course, neglects to mention the task of creating the tempo track, and a myriad of other things that go in to realizing a performance.
Hopefully, this all makes sense. I'm typing this quickly as my baby is getting ready to feed, so I apologize if there's anything in here that's incoherent.
Dave