That is literally not possible.
If you want the normal map to have any meaning, you need to have light information.
Having encoded normals gives you nothing in this context.
Actually you can probably do this sort of stuff in 3ds Max, given that you have the model, but it will essentially be the same as below - generating a light map and multiplying it with the diffuse (also known as baking).
Instead, what you can do is generate a light map, given some constant light, which would result in a black and white gradient, which you can then multiply with the diffuse map in any 2D authoring program (Photoshop, GIMP, etc.)
I don't think it will look good in WC3, but you can try.
Now, if you want something that you can give light information and it will spit said light map, without requiring the model, I don't think anything like that exists.
It's very simple to make, but I don't think anyone has any reasons to do this sort of thing.
Essentially you need to define a directional light (or multiple ones), and then just do the usual dot product between the per-pixel normals and the light direction.
You can probably write something for this in JavaScript in about 10 minutes (if you are interested, I might do it).