As Loren stated, you can't model this directly in CAESAR II. You do realize this isn't a "Code Fitting", so any SIFs computed using Code Equations should be questioned. The elbow flexibility won't be exactly right either.

1) Average properties is one approximation.
2) You could also model two bends (say 2 45s instead of 1 90) and change the properties on the 2nd. This is more difficult because you have to determine the two tangent points.