Secant approximation methods for convex optimization