It's not that easy. I mean, just sending a message could entail one of objc_msgSend, obj_msgSend_stret or objc_msgSend_fpret, and which one you want to use is not always as clear-cut as you might prefer (http://www.sealiesoftware.com/blog/archive/2008/10/30/objc_e...). And then there's the hassle of defining classes dynamically, which entails six function calls just to create a class with one empty method, and one of those essential functions is not actually provided for you AFAIK unless you're using Apple's compiler where it's a builtin, so you'll need to write it yourself.
It's not ridiculously hard, but it's somewhat less easy than normal C FFI.
The other hard part about it is that Haskell's FFI doesn't support variadic functions in a friendly way, which makes argument passing that much trickier.
If you just need to call variadic functions with a specific parameter list determined at compile time, that's easy with the existing FFI; just write a prototype for the function for any given set of arguments you want to call it with. (For instance, you can have an FFI binding printf_int_float :: CString -> CInt -> CFloat -> IO CInt .)
If you really need to construct calls at runtime and you don't know the argument list at compile time, there's a Haskell binding to libffi. However, you generally only need that if you're writing a language interpreter or similar.
There is a similar problem with objc_msgSend in C: Methods with certain argument types can't be called with a normal objc_msgSend(). Instead, you need to create a specifically typed function pointer and assign objc_msgSend to it. I think a similar solution would be necessary with Haskell.
I'd appreciate seeing a link to an example of this. I put a decent amount of work (a few years ago) into writing some bindings in Haskell to the objc runtime, but ended up shelving it until I could come up with a satisfactory solution to the variadic function dilemma.
It's not ridiculously hard, but it's somewhat less easy than normal C FFI.