Hi Mathieu,
It depends on your cross section data...
You could join the points of adjoining cross-sections with lines, re-sample these lines to some appropriate interval and then interpolate values at the resampled points.
Then use those points as source for an isotropic interpolation to your mesh.
Another approach is to create triangles between the points of adjacent cross-sections, save these to a t3s file and then use "Tools->MapObject" to map from your triangulated cross-section mesh to your computational mesh. The interpolation in this case will be linear on each triangle.
As for the curves...
If your cross-section density is high enough to represent the curve then that's what you'll get. If not... well, it's generally unwise to invent data. But if you really insist, you could use splines instead of straight lines to join the cross-sections.
Hope this helps... Martin