From my side I'm using a real time sequencer which calls the application functions.
There is tasks groups every 10,20,50,100,250,500 and 1000 ms. The routine has a processor load monitoring routine which stores the remaining time of each path and the whole remaining time. This system is excellent to manage the tasks when you don't need absolute real time capabilities.
Example : UART.
The UART is interrupt driven, but the function which reads the buffer is called every 20ms which is suffisant and is 0.5ms long on a 89c668 with 20MHZ crystal.

A such sequencer makes you to write correctly the code, avoiding loops. When a function has to wait for an event, then you leave the function and you will test the event occurance next path.

The code size is about 500 bytes with Keil.

Regards
Stephane