Most of the "tenses" of the modern Romance languages are inherited directly from Latin, and so have had plenty of time to accumulate irregularities (and indeed many existed even within Latin).
The two exceptions are the future and conditional (in languages with them, they are notably absent in Sardinian and Balkan Romance like Romanian).
These tenses were originally formed from the infinitive, being followed by a fully conjugated form of the verb habeō (which otherwise survives into Portuguese as haver).
The future tense is formed with the present tense of habeō, and the conditional is formed with its imperfect.
I.e. The forms quoted in the question continue the following constructions in Proto-Romance:
I will think: pensarei < pēnsāre habeō
He will think: pensará < pēnsāre habet
We think: pensaremos < pēnsāre habēmus
They will think: pensarão < pēnsāre habent
Portuguese (and Galician) actually preserve this earlier stage better than most other Romance languages, because in formal language pronominal object clitics can appear between the infinitive stem and the ending (so-called "mesoclisis").
E.g. Formal Portuguese dar-mo-ás (regularly contracted from what would otherwise be dar-me-lo-ás) "you will give me it" vs Colloquial Portuguese & Spanish me lo darás.
This mesoclisis reflects the typical position of pronominal object clitics across Romance, of immediately preceding a finite verb (non-finite verbs can frequently take clitics either before or after them) if and only if the ending is still seen as a separate (albeit cliticised) finite verb, whilst the colloquial form fits the pattern if and only if the entire infinitive + habeō unit is taken to constitute a single finite verb.
Note also, that as part of the process of univerbation, the form of habeō used has been reduced, with the -b- typically dropping out, even in forms where it has typically been retained in the uncliticised form of the verb. In particular, the imperfect of habeō, and some forms in the present plural, generally retain the -b- (typically lenited to -v-) when not cliticised in most of Romance (sometimes alongside reduced forms without the -b-).
The irregular futures you cite (which also have irregular conditionals) also come about because of reduction during this univerbation process. In this instance the reduction occurred in the infinitive part.
The Spanish cognates of fazer and dizer (hacer and decir) show the same irregular dropping of the middle consonant of the infinitive in the future & conditional. This consonant was already lost in the infinitive itself in most of the rest of Romance.
Trazer is somewhat more interesting, because it continues Latin trahō, meaning that the z in the Portuguese is unetymological (likely inserted due to analogy from the perfect participle tractus). Traer (attested in Old Galician-Portuguese, and continuing in Galician) is the expected form.