You can define a convolution iteratively, as f\*g\*h = f\*(g\*h), etc. Or you could can extend the formula by summing over more index variables. So if you define

(g\*h)(n) = sum k=-∞..∞ g(k)h(n-k)

You could extend that as

(f\*g\*h)(n) = sum k=-∞..∞,j=-∞..∞ f(k)g(j)h(n-k-j)

You can continue in that fashion for more variables.

Or if you're asking how to compute it as a practical question, that depends. There's a lot of theory that might be relevant. Is this for a specific problem?