Instruments based on the magneto-optical Kerr effect are routinely used to probe surface magnetic properties. These tools rely on the characterization of the polarization state of reflected light from the sample to collect information on its magnetization. Here, we present a theoretical optimization of common setups based on the magneto-optical Kerr effect. A detection scheme based on a simple analyzer and photodetector and one made from a polarizing beam splitter and balanced photodetectors are considered. The effect of including a photoelastic modulator (PEM) and a lock-in amplifier to detect the signal at harmonics of the modulating frequency is studied. Jones formalism is used to derive general expressions that link the intensity of the measured signal to the magneto-optical Fresnel reflection coefficients for any orientation of the polarizing optical components. Optimal configurations are then defined as those that allow measuring the Kerr rotation and ellipticity while minimizing nonmagnetic contributions from the diagonal Fresnel coefficients in order to improve the signal-to-noise ratio (SNR). The expressions show that with the PEM, setups based on polarizing beam splitters inherently offer a twofold higher signal than commonly used analyzers, and the experimental results confirm that the SNR is improved by more than 150%. Furthermore, we find that while all proposed detection schemes measure Kerr effects, only those with polarizing beam splitters allow measuring the Kerr rotation directly when no modulator is included. This accommodates, for instance, time-resolved measurements at relatively low laser pulse repetition rates. Ultrafast demagnetization measurements are presented as an example of such applications.