Thanks for the feedback, Franck.
The Mesh Deformer might then not the right one for this, except with a lot of weighting and many more of them, if that works at all. Similar to that idea would be the Spline Deformer, where the initial Spline defines the starting point, and the second spline (copy) defines the deformations.
So, why do I think this is not a directions to go here?
The way you describe it, means that the face of the character will be massively deformed. My idea about it, while using joints, one must have a well designed “weight” distribution for each and every joint. Each joint or group of them might go as far as that the whole head is affected. If not everything will move dynamically at the same time, the object {head} will feel folded, scrambled or just fall apart. In the moment you allow for each joint to have influence on the whole mesh, each joint has less influence by itself, as it shares this influence [max 100%].
Why would I think it could work with the Mesh Deformer?
I believe there is a sweet spot, when the base mesh of the object is relatively low, and becomes Subdivided. This would prevent the early folding of the surface. The Cage for the mesh Deformer should be low res naturally, and always handled with the Magnet Tool. The Magnet Tool with a nice falloff will provide a missing overall weighting of the Deformer, but allows for a result that might works.
The ideas to work with Joints and Splines, could work in the same way, by affecting all Splines at once! So there is a movement of all joints, more or less.
Another idea would be the Camera Deformer, as the points of it can be run by a Pose-Morph or even by XPresso. Note: my only critique of this deformer is, that it is very fragile. In the moment one change the aspect ratio of the Render Settings, things change. As I work more often than not in Cinemascope [1:2.40), I notice that, when I changed from UHD to Film settings. Not nice, so keep that in mind.
If I had to do such thing, I would use the Pose Morph in a heart beat. I love to store information rather than to have anything live with the danger to overwrite something I liked and notice it too late.
Who cares how many sliders are in the Pose Morph on the end, what really counts is how easily one can go from one stage to the next stage of an expression. Having said that, and thinking about the “Driver Tag” as base idea, one could also have a Spline over that face and connect point positions to these Sliders, one or several, and via Ranger Mapper fine tune the mix. Not the Driver Tag, but a self build option.
From what I got from the Masters in that area, Rigging means, to have a fast access to many expression, best practice here, one click to get what you want. The simpler the animation work is, the harder the rigging work might have been to get there. Which brings the “Interaction Tag” in to the game perhaps.
Or: Animation Clips with expression, PLA, and with its transitions… But this leaves out the interaction.
Pose Morph and Live changes , well, perhaps this might get you a step further: I builded a Pose-Morph with a spline [green]. Then added a “Live Spline”, to be changed at any time [yellow]. Both were then connected via XPresso>Mix to the resulting Spline [blue]. Just an idea, and very simplified to showcase the idea. Scene file and screenshot, see below.
All the best