A gaussian mixture is a universal density approximator, so you can definitely approximate a lognormal with such a model.