Agreed it's better using the physics because it does a lot for you, but it never hurts to lean something new.
The formula that correlates velocity to angular velocity is:
displacement x velocity = angular_velocity
Where "displacement" is a vector from the pivot to the point of collision and "x" is a vector cross product. "w" will be in radians/sec so we'll need to convert it to degrees/sec by multiplying by 180/pi.
(displacement x velocity)*180/pi = angular_velocity
So to finish it up the displacement vector can be calculated with:
dispX = collisionX - pivotX
dispY = collisionY - pivotY
Also a 2d vector cross product can be defined as:
v1 x v2 = v1x*v2y - v1y*v2x
So them combining those you get a lovely formula like this:
((collisionX - pivotX)*vy - (collisionY - pivotY)*vx)*180/pi = angular_velocity
"vx" and "vy" are the x and y components of velocity. Depending on the behavior used you can also get vx and vy from speed and angle_of_motion:
vx = speed * cos(angle_of_motion)
vy = speed * sin(angle_of_motion)
So to use the formula you could make a collision event when a bullet collides with sprite, and use the bullet's position for collisionX and collisionY.
Now all we have is a way to convert linear motion to angular but it's not enough for it to look realistic. For that you'll need to use momentum and kinetic energy to calculate a collision response. There's info on it elsewhere but the formulas get kind of lengthy and really it's one of the main reasons it's much easier to use an existing physics engine.