Learning More with Less: Self-Supervised Approaches for Low-Resource Speech Emotion Recognition