There are many ways to do this.  Google "FM demodulator" for some ideas.

Designing stable band pass filters with only 50-100 kHz bandwidth at 432.95 and 433.05 MHz is not trivial (Q factor is 5000 to 10000), and may not be the best option.

A common approach in receiver design is to mix the signal with a local oscillator and shift the frequency of the carrier to a lower intermediate frequency, such as 10.7 MHz or 455 kHz.  Narrow band pass filters are now much easier to design as the fractional bandwidth is much higher, so component tolerances less stringent and Q factor can be lower.  If you choose a commonly used IF frequency, there are are likely to be many off-the-shelf demodulators available.

An alternative to your twin BPF discriminator is to use a phase locked loop to track the input signal.  The drive to the voltage controlled oscillator will follow the data modulation.

https://en.wikipedia.org/wiki/Intermediate_frequency#Commonly_used_intermediate_frequencies

http://www.analog.com/en/products/clock-and-timing/plls/integer-n-plls/ad9901.html#product-overview

http://www.radio-electronics.com/info/rf-technology-design/fm-reception/fm-pll-detector-demodulator-demodulation.php

http://www.linear.com/product/LT5512

http://www.analog.com/en/products/rf-microwave/mixers/rf-mixers-w-integrated-los.html

Consider mixing down to 10.7 MHz and a quadrature FM discriminator. The PLL demodulator is a good choice. However, if low cost is desired a super regenerative receiver with a proper quench oscillator may provide the  demodulation function. Search super regenerative detectors. 

Dear Prof. Alan Robinson and Prof. Alan Victor, 

I am already considering the down conversion of the received RF signal to IF band. 

I think several MHz or some KHz should be OK. 

Is it OK to use gilbert cell with LO (Injection Locked Ring Oscillator) ?

Should it work if I am trying to have all of these on-chip for SoC design ?

A single conversion superheterodyne is a fine choice. DC receivers are also attractive although they bring their own set of unique problems. Which choice depends on how clever you can be with the SoC architectures you have available. Choose an IF that is commensurate with reasonable image rejection, minimum spurious response, etc... There is a wide choice of mixer architectures. A simple differential cell will work fine. You do not specify the receiver quality you require such as noise figure, desired bandwidth, dynamic range, spurious responses and all the other alphabet soup acronyms associated with all "receivers". If receiver stability and hence reliable decode of the FSK data is desired, you may ask yourself is the ILO sufficient. Locked to a stable source off chip may be fine.