Ok... what you need to determine is what exactly your mutants should be, i.e. which parts are the bits you want deleted?
Presumably these are significant deletions (20+bp) of the gene of interest. If this is the case, the 5o primer will be the first 20 or so bp of the 5' end of the gene, the 3o will be the reverse complement of the last 20 bp of the gene. The inside primers will be dependent on the location of your deletion, but using the 5o and 3i primers will amplify a fragment that stretches from 5o to 3i and the 5i will amplify in partnership with 3o producing two separate fragments in two separate reactions. These two fragments will flank (be on either side of) your target mutation.
In the protocol you posted, the yellow region is the bit being deleted, so as you can see they have primers that are just on the outside of the yellow region labeled 3i and 5i respectively Their example is a little bit confusing because the OmcA gene appears to be transcribed in the opposite direction to how genes are normally depicted, which is why their 3' and 5' ends are switched.
I don't know how they generated the "tag" sequences, but it looks like they have just used random sequences. Note that the tags will have to be multiples of 3 bases (i.e. codons) in length. For example, they could be 18 or 21 bp, but not 20 as this would alter the reading frame once ligated together. Unless you want to later insert some sort of marker, you don't have to worry about putting restriction sites into the tags. It is not usually a good idea to insert markers into coding sequences anyway as this might well interfere with unique functions of each end of the protein.