Pathfinding only needs a start location and a destination to work... so yes this should be possible.
What you need to do is pick a destination which is directly away from the danger and a distance proportional to how much danger the deer is feeling.
So if your deer is at dx,dy and your wolf is at wx,wy then you want to pick a destination zx,zy as:
closeness = distance(dx, dy, wx, wy)
if closeness <= maximumVisionOfDeer
howFarToRun = maximumFarToRun / closeness
dangerAngle = angle(dx, dy, wx, wy)
safeAngle = dangerAngle + 180
zx = dx + cos(safeAngle) * howFarToRun
zy = dy + sin(safeAngle) * howFarToRun
endif
Then you can pathfind from dx,dy to zx,zy.
Note: maximumFarToRun is the maximum distance the deer will run if the wolf is right next to it (closeness = 1), if your wolf can get even closer than 1 (e.g. 0.1) then this won't work the same and you might want to limit the values into the equation.
Note: if zx,zy is inside a tree, you either need a pathfinding algorithm that will go as close as it can (and not just fail), or you should jiggle the destination a bit with a random offset until it is not in a tree.
There are tons of other things you could add, but this will work to get you started I think.
This is very hard for me to understand, do you think a cap.x would be possible?