Single-molecule localization microscopy has become a prominent approach to study structural and dynamic arrangements of nanometric objects well beyond the diffraction limit. To maximize localization precision, high numerical aperture objectives must be used; however, this inherently strongly limits the depth-of-field (DoF) of the microscope images. In this work, we present a framework inspired by "optical co-design" to optimize and benchmark phase masks, which, when placed in the exit pupil of the microscope objective, can extend the DoF in the realistic context of single fluorescent molecule detection. Using the Cramér-Rao bound (CRB) on localization accuracy as a criterion, we optimize annular binary phase masks for various DoF ranges, compare them to Incoherently Partitioned Pupil masks and show that they significantly extend the DoF of single-molecule localization microscopes. In particular we propose different designs including a simple and easy-to-realize two-ring binary mask to extend the DoF. Moreover, we demonstrate that a simple maximum likelihood-based localization algorithm can reach the localization accuracy predicted by the CRB. The framework developed in this paper is based on an explicit and general information theoretic criterion, and can thus be used as an engineering tool to optimize and compare any type of DoF-enhancing phase mask in high resolution microscopy on a quantitative basis.